Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

BCCN/BFNT Sonderseminar

Monday, 25.10.2010 17 c.t.

Benchmarking the transfer from reflexive to predictive information

by Dr. Bernd Porr
from Department of Electronics & Electrical Engineering, University of Glasgow, UK

Contact person: Florentin Wörgötter


Seminarraum Haus 2, 4. Stock (Bunsenstr.)


In the past we have shown the learning can turn a reflexive system into a proactive system by replacing the reflex by a predictive (re)action. Learning rules such as ICO or ISO learning have been employed successfully to achieve this goal. However learning is not perfect so that usually a system still has to rely on its reflexes once in a while. Also learning might employ a high number of predictive inputs which will all contribute more or less (successfully) to the predictive action. To reflect these two issues we have developed a new performance measure which reflects how successful the transfer from reflex to predictive action has been and how much different predictive inputs contribute to this success. The application of this measure will be shown on the well known model of the driving robot which has to follow a line and learns to predicts curves.

back to overview